pyshmem documentation

pyshmem provides named shared-memory streams for CPU NumPy arrays and optional CUDA-backed PyTorch pipelines. The library is designed for applications that need low-latency data exchange between processes while preserving a simple, uniform API.

User Guide

• Overview
  • Core capabilities
  • Public API
• Installation
  • Base install
  • Optional dependencies
  • Requirements
• Usage
  • CPU stream quickstart
  • GPU stream quickstart
  • Reading modes
  • Waiting for the next write
  • Lifecycle
• GPU Streams
  • Creating a GPU stream
  • CPU mirroring modes
  • Opening GPU streams
• Platform Notes
  • Linux and macOS
  • Windows
• Benchmarks
  • Running benchmarks locally
  • GPU pipeline shapes
  • Environment knobs
• API Reference
  • Top-level package
  • SharedMemory
  • Factory functions
• Release and Publishing
  • PyPI publishing
  • Read the Docs
  • Before tagging a release

Indices and tables

• Index

• Module Index

• Search Page